Get the weekly radar
One short email with the freshest products, every week.
Agent Caly runs an AI task at the time you set, just by inviting it to a calendar event.
Grounded in available product and source data
If you can invite a coworker to a meeting, the same motion works for an AI agent — Agent Caly is built to schedule AI agent tasks from Google Calendar, with no separate app or dashboard needed to assign the work in the first place.
Each agent gets its own email address the moment it's created, and adding that address as a guest on any calendar event, with the task written into the event description, is the entire assignment step. At the event's start time, the agent researches the web, completes the task, and posts a summary back to the user's dashboard — and making the event recurring turns a one-off task into a standing schedule with no extra configuration.
Model choice stays open rather than locked to one provider: any model from Anthropic, OpenAI, or OpenRouter can run a given agent, switchable anytime, with new releases usable the day they launch since the model is just named at creation time. Agents reach further than a single calendar too, connecting to hundreds of tools — Gmail, Slack, Notion, GitHub, HubSpot, and more — each authorized through that service's own login and revocable at any point.
On the obvious trust question, the site doesn't sidestep it: using an AI agent carries real risk of a misread task or an inaccurate result, so the stated guidance is to treat output as a draft, start with minimum access and low-stakes tasks, and expand permissions only once an agent's behavior earns it. A specific spoofing protection is also named directly — the real sender of a calendar invite is checked, not the organizer label shown on the event, so an agent can't be tricked into acting on someone else's behalf.
Not really — the whole assignment step is inviting the agent's own email address as a guest on a Google Calendar event and writing the task in the description, using an interaction most people already know from inviting a coworker.
No — the site states it checks the actual sender of the invite, not the organizer name displayed on the event, specifically to prevent an agent from being spoofed into acting on behalf of someone who isn't its owner.
The site is direct about this — treat what an agent produces as a draft rather than a finished result, especially for anything costly or hard to undo, and expand what an agent can access only after its behavior has earned that trust.
No — a different model from Anthropic, OpenAI, or OpenRouter can be chosen per agent, and switching is possible anytime, with new model releases usable as soon as they launch.
Yes — making the underlying calendar event recurring is enough; the agent then runs on that same cadence automatically, whether that's every weekday, weekly, or a custom schedule.
